2 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
3 * Copyright (c) 2013 Clément Bœsch <u pkh me>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License along
18 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * Simple post processing filter
26 * This implementation is based on an algorithm described in
27 * "Aria Nosratinia Embedded Post-Processing for
28 * Enhancement of Compressed Images (1999)"
30 * Originally written by Michael Niedermayer for the MPlayer project, and
31 * ported by Clément Bœsch for FFmpeg.
34 #include "libavutil/avassert.h"
35 #include "libavutil/imgutils.h"
36 #include "libavutil/mem_internal.h"
37 #include "libavutil/opt.h"
38 #include "libavutil/pixdesc.h"
48 #if FF_API_CHILD_CLASS_NEXT
49 static const AVClass *child_class_next(const AVClass *prev)
51 return prev ? NULL : avcodec_dct_get_class();
55 static const AVClass *child_class_iterate(void **iter)
57 const AVClass *c = *iter ? NULL : avcodec_dct_get_class();
58 *iter = (void*)(uintptr_t)c;
62 static void *child_next(void *obj, void *prev)
65 return prev ? NULL : s->dct;
68 #define OFFSET(x) offsetof(SPPContext, x)
69 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
70 #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
71 static const AVOption spp_options[] = {
72 { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, TFLAGS },
73 { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
74 { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" },
75 { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
76 { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
77 { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, FLAGS },
81 static const AVClass spp_class = {
83 .item_name = av_default_item_name,
84 .option = spp_options,
85 .version = LIBAVUTIL_VERSION_INT,
86 .category = AV_CLASS_CATEGORY_FILTER,
87 #if FF_API_CHILD_CLASS_NEXT
88 .child_class_next = child_class_next,
90 .child_class_iterate = child_class_iterate,
91 .child_next = child_next,
94 // XXX: share between filters?
95 DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = {
96 { 0, 48, 12, 60, 3, 51, 15, 63 },
97 { 32, 16, 44, 28, 35, 19, 47, 31 },
98 { 8, 56, 4, 52, 11, 59, 7, 55 },
99 { 40, 24, 36, 20, 43, 27, 39, 23 },
100 { 2, 50, 14, 62, 1, 49, 13, 61 },
101 { 34, 18, 46, 30, 33, 17, 45, 29 },
102 { 10, 58, 6, 54, 9, 57, 5, 53 },
103 { 42, 26, 38, 22, 41, 25, 37, 21 },
106 static const uint8_t offset[127][2] = {
108 {0,0}, {4,4}, // quality = 1
109 {0,0}, {2,2}, {6,4}, {4,6}, // quality = 2
110 {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3
112 {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4
113 {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
115 {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5
116 {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
117 {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
118 {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
120 {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6
121 {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
122 {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
123 {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
124 {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
125 {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
126 {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
127 {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
130 static void hardthresh_c(int16_t dst[64], const int16_t src[64],
131 int qp, const uint8_t *permutation)
134 int bias = 0; // FIXME
136 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
137 unsigned threshold2 = threshold1 << 1;
139 memset(dst, 0, 64 * sizeof(dst[0]));
140 dst[0] = (src[0] + 4) >> 3;
142 for (i = 1; i < 64; i++) {
144 if (((unsigned)(level + threshold1)) > threshold2) {
145 const int j = permutation[i];
146 dst[j] = (level + 4) >> 3;
151 static void softthresh_c(int16_t dst[64], const int16_t src[64],
152 int qp, const uint8_t *permutation)
155 int bias = 0; //FIXME
157 unsigned threshold1 = qp * ((1<<4) - bias) - 1;
158 unsigned threshold2 = threshold1 << 1;
160 memset(dst, 0, 64 * sizeof(dst[0]));
161 dst[0] = (src[0] + 4) >> 3;
163 for (i = 1; i < 64; i++) {
165 if (((unsigned)(level + threshold1)) > threshold2) {
166 const int j = permutation[i];
167 if (level > 0) dst[j] = (level - threshold1 + 4) >> 3;
168 else dst[j] = (level + threshold1 + 4) >> 3;
173 static void store_slice_c(uint8_t *dst, const int16_t *src,
174 int dst_linesize, int src_linesize,
175 int width, int height, int log2_scale,
176 const uint8_t dither[8][8])
180 #define STORE(pos) do { \
181 temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6; \
183 temp = ~(temp >> 31); \
184 dst[x + y*dst_linesize + pos] = temp; \
187 for (y = 0; y < height; y++) {
188 const uint8_t *d = dither[y];
189 for (x = 0; x < width; x += 8) {
203 static void store_slice16_c(uint16_t *dst, const int16_t *src,
204 int dst_linesize, int src_linesize,
205 int width, int height, int log2_scale,
206 const uint8_t dither[8][8], int depth)
209 unsigned int mask = -1<<depth;
211 #define STORE16(pos) do { \
212 temp = ((src[x + y*src_linesize + pos] << log2_scale) + (d[pos]>>1)) >> 5; \
214 temp = ~(temp >> 31); \
215 dst[x + y*dst_linesize + pos] = temp; \
218 for (y = 0; y < height; y++) {
219 const uint8_t *d = dither[y];
220 for (x = 0; x < width; x += 8) {
234 static inline void add_block(uint16_t *dst, int linesize, const int16_t block[64])
238 for (y = 0; y < 8; y++) {
239 dst[0 + y*linesize] += block[0 + y*8];
240 dst[1 + y*linesize] += block[1 + y*8];
241 dst[2 + y*linesize] += block[2 + y*8];
242 dst[3 + y*linesize] += block[3 + y*8];
243 dst[4 + y*linesize] += block[4 + y*8];
244 dst[5 + y*linesize] += block[5 + y*8];
245 dst[6 + y*linesize] += block[6 + y*8];
246 dst[7 + y*linesize] += block[7 + y*8];
250 static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
251 int dst_linesize, int src_linesize, int width, int height,
252 const uint8_t *qp_table, int qp_stride, int is_luma, int depth)
255 const int count = 1 << p->log2_count;
256 const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16);
257 DECLARE_ALIGNED(16, uint64_t, block_align)[32];
258 int16_t *block = (int16_t *)block_align;
259 int16_t *block2 = (int16_t *)(block_align + 16);
260 uint16_t *psrc16 = (uint16_t*)p->src;
261 const int sample_bytes = (depth+7) / 8;
263 for (y = 0; y < height; y++) {
264 int index = 8 + 8*linesize + y*linesize;
265 memcpy(p->src + index*sample_bytes, src + y*src_linesize, width*sample_bytes);
266 if (sample_bytes == 1) {
267 for (x = 0; x < 8; x++) {
268 p->src[index - x - 1] = p->src[index + x ];
269 p->src[index + width + x ] = p->src[index + width - x - 1];
272 for (x = 0; x < 8; x++) {
273 psrc16[index - x - 1] = psrc16[index + x ];
274 psrc16[index + width + x ] = psrc16[index + width - x - 1];
278 for (y = 0; y < 8; y++) {
279 memcpy(p->src + ( 7-y)*linesize * sample_bytes, p->src + ( y+8)*linesize * sample_bytes, linesize * sample_bytes);
280 memcpy(p->src + (height+8+y)*linesize * sample_bytes, p->src + (height-y+7)*linesize * sample_bytes, linesize * sample_bytes);
283 for (y = 0; y < height + 8; y += 8) {
284 memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp));
285 for (x = 0; x < width + 8; x += 8) {
291 const int qps = 3 + is_luma;
292 qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
293 qp = FFMAX(1, ff_norm_qscale(qp, p->qscale_type));
295 for (i = 0; i < count; i++) {
296 const int x1 = x + offset[i + count - 1][0];
297 const int y1 = y + offset[i + count - 1][1];
298 const int index = x1 + y1*linesize;
299 p->dct->get_pixels_unaligned(block, p->src + sample_bytes*index, sample_bytes*linesize);
301 p->requantize(block2, block, qp, p->dct->idct_permutation);
302 p->dct->idct(block2);
303 add_block(p->temp + index, linesize, block2);
307 if (sample_bytes == 1) {
308 p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
309 dst_linesize, linesize, width,
310 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
313 store_slice16_c((uint16_t*)(dst + (y - 8) * dst_linesize), p->temp + 8 + y*linesize,
314 dst_linesize/2, linesize, width,
315 FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
322 static int query_formats(AVFilterContext *ctx)
324 static const enum AVPixelFormat pix_fmts[] = {
325 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P,
326 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV411P,
327 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV440P,
328 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
329 AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
330 AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV422P10,
331 AV_PIX_FMT_YUV420P10,
332 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV422P9,
341 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
343 return AVERROR(ENOMEM);
344 return ff_set_common_formats(ctx, fmts_list);
347 static int config_input(AVFilterLink *inlink)
349 SPPContext *s = inlink->dst->priv;
350 const int h = FFALIGN(inlink->h + 16, 16);
351 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
352 const int bps = desc->comp[0].depth;
354 av_opt_set_int(s->dct, "bits_per_sample", bps, 0);
355 avcodec_dct_init(s->dct);
360 s->hsub = desc->log2_chroma_w;
361 s->vsub = desc->log2_chroma_h;
362 s->temp_linesize = FFALIGN(inlink->w + 16, 16);
363 s->temp = av_malloc_array(s->temp_linesize, h * sizeof(*s->temp));
364 s->src = av_malloc_array(s->temp_linesize, h * sizeof(*s->src) * 2);
366 if (!s->temp || !s->src)
367 return AVERROR(ENOMEM);
371 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
373 AVFilterContext *ctx = inlink->dst;
374 SPPContext *s = ctx->priv;
375 AVFilterLink *outlink = ctx->outputs[0];
378 const int8_t *qp_table = NULL;
379 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
380 const int depth = desc->comp[0].depth;
382 /* if we are not in a constant user quantizer mode and we don't want to use
383 * the quantizers from the B-frames (B-frames often have a higher QP), we
384 * need to save the qp table from the last non B-frame; this is what the
385 * following code block does */
387 qp_table = av_frame_get_qp_table(in, &qp_stride, &s->qscale_type);
389 if (qp_table && !s->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
392 /* if the qp stride is not set, it means the QP are only defined on
395 w = AV_CEIL_RSHIFT(inlink->w, 4);
399 h = AV_CEIL_RSHIFT(inlink->h, 4);
402 if (w * h > s->non_b_qp_alloc_size) {
403 int ret = av_reallocp_array(&s->non_b_qp_table, w, h);
405 s->non_b_qp_alloc_size = 0;
408 s->non_b_qp_alloc_size = w * h;
411 av_assert0(w * h <= s->non_b_qp_alloc_size);
412 memcpy(s->non_b_qp_table, qp_table, w * h);
416 if (s->log2_count && !ctx->is_disabled) {
417 if (!s->use_bframe_qp && s->non_b_qp_table)
418 qp_table = s->non_b_qp_table;
420 if (qp_table || s->qp) {
421 const int cw = AV_CEIL_RSHIFT(inlink->w, s->hsub);
422 const int ch = AV_CEIL_RSHIFT(inlink->h, s->vsub);
424 /* get a new frame if in-place is not possible or if the dimensions
425 * are not multiple of 8 */
426 if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
427 const int aligned_w = FFALIGN(inlink->w, 8);
428 const int aligned_h = FFALIGN(inlink->h, 8);
430 out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
433 return AVERROR(ENOMEM);
435 av_frame_copy_props(out, in);
436 out->width = in->width;
437 out->height = in->height;
440 filter(s, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1, depth);
443 filter(s, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0, depth);
444 filter(s, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0, depth);
452 av_image_copy_plane(out->data[3], out->linesize[3],
453 in ->data[3], in ->linesize[3],
454 inlink->w, inlink->h);
457 return ff_filter_frame(outlink, out);
460 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
461 char *res, int res_len, int flags)
463 SPPContext *s = ctx->priv;
465 if (!strcmp(cmd, "level") || !strcmp(cmd, "quality")) {
466 if (!strcmp(args, "max"))
467 s->log2_count = MAX_LEVEL;
469 s->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
472 return AVERROR(ENOSYS);
475 static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts)
477 SPPContext *s = ctx->priv;
480 s->dct = avcodec_dct_alloc();
482 return AVERROR(ENOMEM);
485 AVDictionaryEntry *e = NULL;
487 while ((e = av_dict_get(*opts, "", e, AV_DICT_IGNORE_SUFFIX))) {
488 if ((ret = av_opt_set(s->dct, e->key, e->value, 0)) < 0)
494 s->store_slice = store_slice_c;
496 case MODE_HARD: s->requantize = hardthresh_c; break;
497 case MODE_SOFT: s->requantize = softthresh_c; break;
502 static av_cold void uninit(AVFilterContext *ctx)
504 SPPContext *s = ctx->priv;
509 av_freep(&s->non_b_qp_table);
512 static const AVFilterPad spp_inputs[] = {
515 .type = AVMEDIA_TYPE_VIDEO,
516 .config_props = config_input,
517 .filter_frame = filter_frame,
522 static const AVFilterPad spp_outputs[] = {
525 .type = AVMEDIA_TYPE_VIDEO,
530 AVFilter ff_vf_spp = {
532 .description = NULL_IF_CONFIG_SMALL("Apply a simple post processing filter."),
533 .priv_size = sizeof(SPPContext),
534 .init_dict = init_dict,
536 .query_formats = query_formats,
537 .inputs = spp_inputs,
538 .outputs = spp_outputs,
539 .process_command = process_command,
540 .priv_class = &spp_class,
541 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,